Improving Quality of Service Parameter Prediction with Preliminary Outlier Detection and Elimination

Wide-spread real-time applications make it necessary for service providers to guarantee QoS parameters. This requires precise forecast of network traffic. A possible method of the forecast is measuring traffic then analyzing it and fitting model to the measured data, finally predicting the observed parameter using the fitted model. The efficiency of the prediction is decreased by outlying samples (so called outliers) found in the time series data. We developed a new tool that is able to detect and eliminate the outliers from time series data. This tool is capable to handle large sets of time series data fast and efficiently. We also propose a method to predict QoS parameters using the ARIMA (Auto-Regressive Integrated Moving Average) model, which is based on a preliminary detection and elimination of outliers. We have proven that this method increases the efficiency of the prediction significantly by forecasting real measurement data.